Tools
Questions about one object, not one market
The statistics on every artist page describe a market. These answer something narrower: what a specific work is worth, where it should be consigned, and whether it will sell at all — each computed from the 4,679,234 lots that failed to sell as well as the ones that did.
Where to sell expected proceeds by auction house
Median hammer multiplied by panel sell-through, over lots by the same artist whose own printed estimate sat near yours. The house with the highest median and a 40% sell-through returns less, on average, than one a third cheaper that sells everything — which is the trade a prices-realized database cannot show you, because it discarded the lots that failed.
Open to everyone: the ranking, each house's depth and its sell-through. With a Research Pass or a subscription: the median hammer and the expected proceeds. Figures are hammer, gross of the seller's commission. Method →
Find an artist above, then enter the estimate — the same form drives this and the percentile below.
Place a value free, no account
Where a dollar figure sits in an artist's recorded distribution, optionally narrowed to one medium, with the full spread drawn behind it. The spread is the point: it is the argument for why an average cannot value a work, and why the comparables matter more than the median.
Will it sell? sell probability on upcoming lots
A weighted logistic regression on the measured sell-through of the estimate band, the house, the artist, the month and the category, trained only on houses and years whose archives actually record bought-in lots. Validated out of time, and the calibration table is published rather than asserted.
Open to everyone: the probability itself and the model's validation. With a subscription: the factor breakdown — which term moved this lot, and by how much. Method →
Pre-sale forecasts marquee evening sales, scored afterward
Every marquee evening sale gets a card before it happens: expected sell-through, an 80% range on the sold count, and the ten lowest-scored lots. After the sale the card is scored against the recorded result and stays up, right or wrong. No other price database publishes a forward-looking number it then marks itself against.
Open to everyone: the sale forecast, the ten lowest-scored lots, and the model's held-out validation. With a Research Pass or a subscription: every scored lot with its probability and the factor that moved it.
Estimate Reliability League free, quarterly
Houses ranked by how close their estimates land to the hammer, by department and quarter, with estimate width, the shares above high and below low, and the bought-in rate beside them. Panel-gated, and every row prints its sample.
For software API and MCP
Every figure the site computes is available machine-to-machine, and every
derived number returns the lot ids it was computed from — so a valuation can
be defended a year later. Any query can be asked as_of a date,
which is what a date-of-death valuation or a re-run of last quarter's mark
actually needs.
The same tools are available to AI assistants over MCP — one stdlib-only file that turns a key into callable tools, so a model answering a question about an artist can cite the lots it read rather than remembering them. Every competitor of comparable size publishes no API at all, and the largest index publisher in this market blocks AI crawlers outright.